Hmm, maybe I am getting things confused.

It is using xalan/xerces for the processor, and I'm using saxon and
cocoon as well.

Cocoon uses Xerces/Xalan by default. You can't use *both*
Saxon and Xalan at the same time unless you did some
advaced hackery to the Cocoon configuration files.

anyway, the message seems to come from Xerces (1.x). Form
other experience, this could be easily caused by a wrong
encoding declaration. Perhaps it would help to see the
first few lines of the XML input, but all in all it
seems to be more of a case for the cocoon-users list.